Fix 3D materials not updating on texture invalidation

[briansemrau]

See better implementation: #54489

---

Fixes #51361
Fixes #53206
Fixes #53089
Should fix #42821
Might fix #50296
Might fix #52505
Might fix #45746

It's possible there are other issues related to/fixed by this.

Might be relevant for #50327.

Supersedes #52528

2021-10-29.10-18-00.mp4

---

Explanation of the issue here:
#51361 (comment)

This is essentially finishing what #50269 set out to do. Implementation derived from:

godot/servers/rendering/renderer_rd/renderer_canvas_render_rd.cpp

Lines 1378 to 1382 in 0dc809c

	if (!RD::get_singleton()->uniform_set_is_valid(md->uniform_set)) {
	// uniform set may be gone because a dependency was erased. In this case, it will happen
	// if a texture is deleted, so just re-create it.
	storage->material_force_update_textures(material, RendererStorageRD::SHADER_TYPE_2D);
	}

[clayjohn]

I think we will need to add the same fix to particle shaders, sky shaders, and fog shaders.

[briansemrau]

I'll hunt down all missing spots. I noticed I've missed a few.

[briansemrau]

I checked all usage of RendererStorage::material_get_data, so this should cover everything.

[clayjohn]

This is an area of the code I am unfamiliar with, but it looks like reduz wrapped the calls in a check to ensure that material_force_update_texture was only ever called once per frame

godot/servers/rendering/renderer_rd/renderer_canvas_render_rd.cpp

Lines 1376 to 1383 in 0dc809c

	if (md->last_frame != RendererCompositorRD::singleton->get_frame_number()) {
	md->last_frame = RendererCompositorRD::singleton->get_frame_number();
	if (!RD::get_singleton()->uniform_set_is_valid(md->uniform_set)) {
	// uniform set may be gone because a dependency was erased. In this case, it will happen
	// if a texture is deleted, so just re-create it.
	storage->material_force_update_textures(material, RendererStorageRD::SHADER_TYPE_2D);
	}
	}

I'm wondering if it should be the same for the other calls

[briansemrau]

Yup, seems those checks were intended design, as last_frame is unused in a lot of materials. Updated.
The uniform set should only be invalid once per frame, so maybe this isn't a big deal in the end. The only code run every material use would be RID_Alloc::owns, which doesn't look expensive, but redundant nonetheless.

[clayjohn]

Looks good to me! This is something that reduz should have the final say on though.

[briansemrau]

Somehow didn't consider that there's a lot of code repetition here. If this is the preferred solution, it would be good for this to have its own function.

[akien-mga]

Superseded by #54489.